EP2144444A1

EP2144444A1 - HDR video data compression devices and methods

Info

Publication number: EP2144444A1
Application number: EP08012452A
Authority: EP
Inventors: Francesco Banterle; Alessandro Artusi; Kurt Debattista; Patrick Ledda; Alan Chalmers; Gavin John Edwards; Gerhard Bonnet
Original assignee: University of Warwick
Current assignee: University of Warwick
Priority date: 2008-07-10
Filing date: 2008-07-10
Publication date: 2010-01-13
Anticipated expiration: 2028-07-10
Also published as: US9406112B2; WO2010003692A1; ES2389458T3; US20110235720A1; EP2144444B1

Abstract

The invention relates to a method of compressing a stream of video frame data wherein tone mapping functions are determined for video frames of said stream, said tone mapping functions being different from one another for frames relating to different scenes and wherein it is suggested that the tone mapping functions will be altered for frames of the stream relating to the same scene.

Description

The present invention relates to the handling of high quality video data.
The quality of digital images has vastly improved over the past few years. The number of pixels in a picture has increased due to improvements in both cameras and.displays. For example, the resolution of commercial flat panel displays used for television has increased step by step from 640x480 pixels VGA resolution to full high definition having 1920x1080 pixels. This has already brought about a change in the amount of data that needs to be handled to display an image. Furthermore, not only has the spatial resolution been improved, but also the dynamic resolution. For a black and white picture this would mean that a single pixel could represent any of a large number of different shades of grey instead of only being black or white. The overall dynamic range easily compares to the human eye. A human being can distinguish objects both in starlight during night time or in bright sun light, even though on a moonless night objects receive only approximately 1/1.000.000.000 of the illumination they would on a bright sunny day: This corresponds to a dynamic range of 90 db. However, the eye needs time to adjust to different light levels. Thus, the dynamic range of the human eye without adjustment of the pupil is only approximately 30 db. In contrast, the differences between very dark and very light spots in a picture taken by a modern camera system can easily be larger than those the human eye can distinguish without adaption; with a modern camera system it is thus possible to determine fine details in very dark spots even while very bright spots are also present in a picture. The dynamic range of a modern camera system may also easily surpass the dynamic range of a conventional display.
It should also be noted that not only the dynamic range of a picture to be (re-)produced needs amendment but also a correction is necessary due to numerous technological limitations such as limited colour gamut and contrast, limited spatial resolution, usually limited field of view and nontrivial workarounds to achieve stereo capacity. Furthermore, in order to reproduce the correct appearance, it is often necessary to simulate the behaviour of the human visual system. Here, it should be noted that the viewing conditions of an observer, observing either the scene or a display may be completely different. Tone mapping is very helpful in the production of realistic images and several operators have been proposed. However, it is still prohibitive to have a complex tone mapping operator that produces high quality results in real time due to restriction of the data processing capacity, in particular for large high definition pictures. Two main tone mapping operator classes exist, namely tone reproduction curves (TRCs) and tone reproduction operators (TROs). Both will be referred to as tone mapping functions in the present application.
TRC algorithms are efficient because the operation is applied to pixels independently and thus can be performed in parallel using a simple look-up table. In addition, models also exist that are able to capture some important aspects, such as visual adaption. However, TRCs fail to capture the important information on local contrast that could be represented in the spatial context of neighbouring image pixels which is of great importance for the human visual system.
The algorithms based on TROs when compared with TRCs are able to capture the important information on local contrast. Unfortunately, they typically introduce artefacts in some parts of the image, such as dark halos, and are computationally more demanding.
Now, images that have a high dynamic range have a higher pixel depth than the more conventional low dynamic range images. The increased bit depth per pixel presented by HDR-image formats can account for all the dynamic range visible by the human visual system. Both the increased number of pixels in a high definition (HD) picture and the improved dynamic information in a high dynamic range (HDR) picture increase the amount of data relating to a picture. This poses severe technical problems. The data size gives rise to problems when transmitting and/or storing image data. These problems become more severe when video data consisting of a plurality of frames need to be handled instead of single digital images. For high dynamic range still pictures, the above-mentioned tone mapping reduces the dynamic range of a picture in a way that attempts to allow the user to observe all details relevant to the human eye even if this would not be possible in a conventional scene and/or adaption. Reducing the dynamic range may also be used to reduce the amount of data necessary to fully describe the picture. Furthermore, for video streams consisting of a sequence of digital images (frames), methods that allow reduction (compression) of data have been described such as the MPEG-standard.
However, despite the fact that both the high definition as well as high dynamic range information may be compressed, the results when applying such conventional compression schemes to a stream of video data are still not satisfying for a user.
High Dynamic Range Images (HDRIs) consume a considerable amount of memory, and efficient compression methods are not easy to implement. A typical low dynamic range image consists of 24 bit-per-pixel (bpp) RGB data that can be compressed considerably. On the other hand, an uncompressed HDR image typically uses 96 bpp for an RGB image. Accordingly, an uncompressed HDR image consumes four times the memory of an uncompressed LDR image.
The first attempt to compress the HDR data was introduced by G. WARD in REAL PIXELS, Graphics Gems, 2:15-31, 1991. It has been suggested that the 96 bpp image is compressed to 32 bpp using a light bit mantissa for each channel and a shared light bit exponent. The resulting RGBE representation however does not cover the full visible colour gamut since it does not allow for negative values. Other formats have been suggested for example in G. W. LARSON: LOGLUV ENCODING FOR FULL-GAMUT, HIGH DYNAMIC RANGE IMAGES, Journal of Graphics Tools, 3(1) :15-31, 1998. Further suggestions, using different data formats can be found in INDUSTRIAL LIGHT & MAGIC, OpenEXR., http://www.openexr.org, 2002 and G. McTAGGERT, C. GREEN and J. MITCHELL: HIGH DYNAMIC RANGE RENDERING IN VALVE'S SOURCE ENGINE, in SIGGRAPH '06: ACM SIGGRAPH 2006 Courses, . Furthermore, several extension to standard compression algorithms have been presented such as for example in G. WARD: JPEG-HDR: A BACKWARDS-COMPATIBLE, HIGH DYNAMIC RANGE EXTENSION TO JPEG, in: CIC 13th: Proceedings of the Thirteenth Color Imaging Conference, The Society for Imaging Science and Technology, 2005; G. WARD : A GENERAL APPROACH TO BACKWARDS-COMPATIBLE DELIVERY OF HIGH DYNAMIC RANGE IMAGES AND VIDEO in: CIC 14th: Proceedings of the Fourteenth Color Imaging Conference, The Society for Imaging Science and Technology, 2006; G. WARD and M. SIMMONS: SUB BAND ENCODING OF HIGH DYNAMIC RANGE IMAGERY in: APGV '04: Proceedings of the 1st Symposium on Applied Perception in Graphics and Visualization, pages 83-90, New York, NY, USA, 2004, ACM Press.
Here, backwards-compatible HDR-JPEG extends the JPEG-standard keeping retro-compatibility. Firstly, an HDR-image is tone mapped using a standard tone reproduction curve and stored as a normal JPEG. Secondly, a sub-band corresponding to HDR in-formation is stored in the "application markers" of the standard for a maximum of 64 Kbytes which is a constraint for encoding high resolution HDR images. HDR-JPEG 2000 is an extension to JPEG 2000 that exploits the support of the standard for 16 bit integer data. With this method, HDR data is transformed into the logarithmic domain, quantized into integers and compressed using standard JPEG 2000 encoding.
While the suggested compression schemes known in the art are helpful for still images, they leave room for improvement and are hardly applicable for HDR videos.
There are two important applications for video data streams providing HDR video data. The first application is movie pictures such as produced by video cameras; the second is graphic applications such as in video games and so forth. Here, there is a particular need to compress the high dynamic range of textures. HDR textures are typically used in interactive applications; due to the size of HDR images they need to be compressed. Methods related to high dynamic range texture compression can either be designed for general purpose graphics hardware or custom made processors that allow decoding of the HDR textures in real time.
The problem of HDR texture compression has been addressed by numerous authors, compare e.g. J. MUNKBERG, P. CLARBERG, J. HASSELGREN, and T. AKENINE-MÖLLER: HIGH DYNAMIC RANGE TEXTURE COMPRESSION FOR GRAPHICS HARDWARE in: ACM Trans. Graph., 25(3):698-706, 2006; K. ROIMELA, T. AARNIO and J. ITARANTA: HIGH DYNAMIC RANGE TEXTURE COMPRESSION in: ACM Trans. Graph., 25(3):707-712, 2006; K. ROIMELA, T. AARNIO and J. ITARANTA; EFFICTENT HIGH DYNAMIC RANGE TEXTURE COMPRESSION in: S13D'08: Proceedings of the 2008 Symposium on Interactive 3D Graphics and Games, pages 207-214, ACM Press, New York, NY, USA, 2008. One problem of the methods described in the prior art is that they require complex hardware.
For HDR videos, HDR-MPEG compression schemes have been proposed in: R. MANTIUK, A. EFREMOV, K. MYSZKOWSKI, and H.-P. SEIDEL: BACKWARD COMPATIBLE HIGH DYNAMIC RANGE MPEG VIDEO COMPRESSION, ACM Trans. Graph., 25(3):713-723, 2006; R. MANTIUK, G. KRAWCZYK, K. MYSZKOWSKI, and H.-P. SEIDEL: PERCEPTION-MOTIVATED HIGH DYNAMIC RANGE VIDEO ENCODING, ACM Trans. Graph., 23(3):733-741, 2004. The suggested schemes can be used as an extension to MPEG-4. As HDR-JPEG backwards-compatible videos are tone mapped, for each frame a reconstruction function is calculated when storing the HDR data. To improve quality, residuals of frames are saved in the video stream. While these algorithms present high quality and high compression ratios, they are not ideally suitable for real time applications since their lack of hardware support results in complex implementations, particularly due to the complex fetching mechanisms required for decoding.
A further approach has been adopted by L. WANG, X. WANG, P-P SLOAN, L-Y WEI, X. TONG and B. GUO: RENDERING FROM COMPRESSED HIGH DYNAMIC RANGE TEXTURES ON PROGRAMMABLE GRAPHICS HARDWARE, I3D '07: Proceedings of the 2007 symposium on Interactive 3D graphics and games, 17-24, ACM Press, New York, 2007. In this paper it has been suggested to separate HDR and LDR parts of the images and to quantize two 8-bit textures compressed using S3TC with their residuals. The reconstruction (decoding) was performed by combining HDR and LDR parts using a simple shader. A more general compression scheme has recently been proposed in S. LEFEBVRE and H. HOPPE: COMPRESSED RANDOM-ACCESS TREES FOR SPATIALLY COHERENT DATA in: Rendering techniques (Proceedings of the Eurographics Symposium on Rendering), Eurographics, 2007. The method described therein relies on a hierarchical data structure that represents spatially coherent graphics data. Despite the good compression the shader is complex and about twenty times slower than a fetch to a compressed texture using S3TC.
It should be noted that hardware solutions for RGBE filtering and so forth have been suggested as well, compare M. KILGARD, P. BROWN and J. LEECH: GLEXT TEXTURE SHARED EXPONENT in: OpenGL Extension, http://www.opengl.org/registryJspecs/EXT/texture_shared_expon ent.txt, 2007 as well as in D. BLYTHE: THE .
It has also already been suggested to use inverse tone mapping for compression. Here, a multi-scale image processing technique for both tone mapping and companding has been proposed by Y. LI, L. SHARAN and E. H. ADELSON: COMPRESSING AND COMPANDING HIGH DYNAMIC RANGE IMAGES WITH SUB BAND ARCHITECTURES in: SIGGRAPH '05: ACM SIGGRAPH 2005 Papers, pages 836-844, New York, NY, USA, 2005, ACM Press. However, the operation is not efficient on current hardware given the fact that the compressed low dynamic range has to be decomposed into sub bands. A compression method based on an inverse tone mapping operator and JPEG was presented in M. OKUDA and N. AD-AMI: RAW IMAGE ENCODING BASED ON POLYNOMIAL APPROXIMATION, IEEE International Conference on Acoustics, Speech and Signal Processing, 2007, ICASSP 2007, pages 15-31, 2007. The tone mapping operators based on the Hill functionals for inverse tone mapping are calculated using minimization techniques and then encoded using JPEG. The residuals are calculated for increased quality and are compressed using wavelets. However, wavelets and DCT decompression are computational expensive to evaluate and do not provide a constant decompression time which is particularly disadvantageous in real time critical applications.
A compression scheme for still images and videos using a tone mapping operator based on a model of a human cones was presented in J.H.V. HATEREN: ENCODING OF HIGH DYNAMIC RANGE VIDEO WITH A MODEL OF HUMAN CONES, ACM Trans. Graph., 25(4):1380-1399, 2006.
The above discussion shows that despite numerous attempts at compressing high dynamic range images, the results achieved thus far are not completely satisfying in view of compression efficiency, quality of result images and computational load. These problems increase when streams of video frame data need to be handled. As long as the stream of video data corresponds to a slide show of only still images, it will be obvious that a tone mapping function used for compression can be altered whenever the still image (and thus in a context of a video stream the scene) changes. However, in a conventional video stream, there are a number of cases where the scene does not change abruptly, for example in cases where a panning, tilting or zooming in of a camera occurs and/or one or a plurality of objects move within a given picture and/or the illumination changes, for example in cases where a room is shown where lights are switched on or off. In cases like this, it is necessary to provide both an efficient compression as well as a compression that does not lead to severe artefacts in a reproduced video.
The present invention aims at providing an improved method of compressing a stream of video frame data as well as devices for implementing the method.
According to a first basic idea of the present invention, a method of compressing a stream of video frame data is suggested wherein tone mapping functions are determined for video frames of said stream, said tone mapping functions being different from one another for frames relating to different scenes and wherein the tone mapping functions will be altered for frames of the stream relating to the same scene.
It has been found that it is clearly advantageous to determine a different tone mapping function not only for every scene but to change the tone mapping function even while the same scene is shown in a video stream. The method of the present invention thus provides a stream of video data which has a significantly better quality compared to a video stream where a change of tone mapping function is only effected for a change of scene. It is possible and preferred to apply the method of compressing the stream to high definition video data, i. e. images that comprise a large number of pixels. Here, the present invention provides advantages as fine spatial details offered by high definition video data can be more clearly identified while the scene lasts.
In the present invention, it is highly preferred to alter the tone mapping function within a scene only gradually. By altering the tone mapping function only gradually, the differences from frame to frame will be small and thus the received image will seem to be more natural. It should be noted that in cases where, without a change of scene, the illumination of a scene is altered for example by switching on or off a light, exceptions from the general rule that the tone mapping function is to be altered gradually only, can be made.
Accordingly, it is possible to evaluate the overall luminance of an image and to allow for larger changes when there is a significant change of luminance from frame to frame. However, it is obvious that changes like this could also be accounted for by handling luminance separately.
The gradual change of the tone mapping function can be achieved by evaluating both a given frame as well as a plurality of previously observed frames and/or frames that will subsequently follow within the data stream. This can be effected either by calculating the average of image data and/or by calculating data derived from original frames such as bilaterally filtered base frames and/or by determining a tone mapping function for any given frame in response to tone mapping functions determined for previous frames.
By relating the current tone mapping function to both a given frame and a number of previously obtained tone mapping functions instead of the frames they are derived from, the amount of data is significantly reduced in comparison to cases where a plurality of video frames has to be evaluated in order to determine one single tone mapping function. Here, it should be noted that once a base frame having reduced spatial resolution is determined from an initial HDR frame using e. g. bilateral filtering, the amount of data that needs to be handled for determining the tone mapping function is significantly reduced, thus allowing even for the calculation of averages and so forth. As the determination of a tone mapping function for a base frame is computationally more simple without degrading image quality, it is highly preferred to separate the base frame information from a given HDR frame prior to determining the tone mapping function and to determine the tone mapping function in response to the base frame information extracted from the initial HDR frame. It is also possible to determine base frame information only for some frames of the video stream; however, it is obviously preferred to determine a base frame for the majority of frames of the initial HDR frames and to operate on this majority of frames in order to determine the tone mapping function. In particular, it is preferred to determine a base frame for every frame of the initial HDR video stream.
It is possible to determine a low dynamic range base frame in response to a base frame extracted from the initial HDR frame and information relating to the tone mapping function for the respective base frame. It should be noted that it is preferred to include information regarding the current tone mapping function used for any given frame in the stream of compressed video data. This can be done in a plurality of ways without significantly increasing the amount of data needed. In particular, it is possible to refer to a number of predefined tone mapping functions and to select one of a number of predefined tone mapping functions for any given frame. This number and any associated parameters can be included in the compressed stream of video frame data. The decoder will usually have access to or comprise a look-up-table for tone mapping functions.
It is possible to predefine every single tone mapping function and to provide information on the tone mapping function for the decoder. In the decoding process, it is also possible to amend an initial tone mapping function so as to directly take into account the particular properties of a device for displaying the actual video frame data. Thus, the look-up-tables for different display devices may relate to different inverse tone mapping functions for a given look-up-table entry. It should be noted that the number of predefined different tone mapping functions can be comparatively small and that between 16 and 512 predefined tone mapping functions should generally be sufficient for encoding and compressing any given HDR video frame data stream.
However, it is also possible to allow the user to define one tone mapping function or a plurality of tone mapping functions as needed. Since any information of a user defined tone mapping function can be included in the compressed stream of video frame data, it is possible to transfer the information of user defined tone mapping functions from the encoder to the decoder.
It should be noted that once the base frame has been determined from the initial high dynamic range frame, it is possible to determine a detail frame in response to both the initial frame and the base frame. The detail frame can preferably be obtained by subtraction of the base frame from the HDR frame and/or by division of the base frame.
It should be noted that generally the detail frame has a sig-nificantly lower dynamic range than the base frame even without dynamic compression. The low dynamic range base frame can be subjected to further compression, in particular to temporal compression schemes as known in the art. It should also be noted that the detail frame can also be compressed further using (conventional) temporal compression schemes. Accordingly, a final frame in the stream of compressed HDR data generally will comprise information on the tone mapping function used for the particular frame, as well as a compressed base frame information and compressed detail frame information. Additional information such as data headers and so forth may obviously be included. When decoding the video stream, the low dynamic range base frame information and the low dynamic range detail frame information can be obtained as known per se in the art. Then, applying the tone mapping information, the initial high dynamic range base frame can be reconstructed and the detail frame information as re-obtained from the temporally compressed detail frame can be added and/or in other ways combined with the uncompressed base frame information to obtain a reproduced HDR frame of high quality. A video stream obtained in this way will provide high quality images while not imposing high computational loads on any graphics hardware processing the initial HDR video frame data.
It should be noted that it is possible and preferred to include in the compressed video data stream inter alia luminance information referring to the overall luminosity of a frame and/or of parts thereof separately from the tone mapping information.
It should be noted that although the present application refers generally to frames and one tone mapping function used for one single frame, it would be possible to use different tone mapping functions for different areas of a given frame without deviating from the disclosure of the present invention. In cases where different tone mapping function information is determined for different areas, a plurality of possibilities exist and it should be noted that any possibility can be applied exclusively or in combination with others. In particular, it is possible to determine a tone mapping function information for a fixed area, for example a fixed rectangle in a given frame. This area may or may not relate to a given object; in particular, it is possible to provide a tone mapping function for moving objects in the initial stream of video frame data. This allows for maintaining tone mapping for any given object moving through a scene. It should be noted that this may require the identification of objects within the video data stream to be compressed. However, since it is possible to identify moving objects within the base frames, the amount of data that needs to be handled can be comparatively small even for high definition pictures. If a plurality of tone mapping functions are determined for one frame, then it is obviously preferred to gradually alter each single tone mapping function for any given area from frame to frame. However, this may not hold for special cases, e. g. when a scene shows a room with a TV set where a movie is shown; now, even if the room remains unchanged otherwise and tone mapping functions for any place in the room other than for the TV set shown change gradually only, there may be large changes in the TV set thus requiring non-gradual changes in the tone mapping function.
The present invention also relates to a video encoder for decoding a stream of video data as described with respect to the methods above. A video encoder can be implemented in software or in hardware, e. g. as ASIC. It will be obvious to the average skilled person how a decoder has to be designed.
Furthermore, it is possible to provide a video camera that stores and/or transmits a video data stream obtained according to a method as described above.
It is possible to provide streams of video data according to the present invention, for example on DVDs or/as downloadable video data streams.
The present invention also relates to a data carrier carrying video data compressed according to the present invention.
The present invention will now be described by way of example only with respect to the drawing wherein it is shown by

Fig. 1: a compression scheme according to the present invention.

According to Fig. 1, a stream of video frame data 1 consisting of high definition high dynamic range video frames is provided. A single HDR frame 2 of the video stream 1 is inputted to a filter 3 which may be implemented by either soft-or hardware and which is used to extract a base frame 4 from the HDR frame 2. The filter 3 may be a bilateral filter.
The base frame 4 is combined in a combining stage with the initial HDR frame, for example by subtracting the base frame 4 from the HDR frame 2 and/or by dividing the HDR frame 2 by the base frame 4 so as to provide a detail frame 5 relating to the fine details of the HDR frame 2. The base frame 4 is further fed to a tone mapping operation stage 6 wherein one of a plurality of - here predefined tone mapping functions stored in a tone mapping look-up-table 7 is selected for the base frame.
The tone mapping function 6 selected for a given base frame 4 is altered to a tone mapping function TMO* with respect to the temporal properties of HDR frames and/or previous tone mapping functions as will be described later on. The tone mapping function TMO* is then applied to the base frame 4 so as to obtain a low dynamic range base frame 8. The low dynamic range base frame 8 and the detail frame 5 are subjected to temporal compression algorithms separately at stages 9 and 10 respectively. The temporal compression schemes may be as known per se in the art and may relate inter alia to MPEG encoding and so forth.
From the temporally compressed detail frames and base frames 9 and 10 and the temporally corrected tone mapping function TMO*, a final frame is constructed. In order to obtain a high quality compressed video stream, it is suggested in the embodiment described to determine the tone mapping function with respect to tone mapping functions previously obtained for previous frames. Thus, the tone mapping function for a given base frame is fed to a temporal property stage 11 which stores the properties of a number of previous tone mapping functions obtained for previous frames of the same scene. This can be done by storing the number of the previous tone mapping functions in a tone mapping function table 7 and/or by storing the entire tone mapping function. It should be noted that a scene change identifier stage may be provided to reset the temporal property stage on any change of scene (not shown). The temporal property stage 11 on receiving an initial tone mapping function as initially determined for a base frame 4 determines whether or not the change in the tone mapping function compared to prior tone mapping function is large or not. It will be obvious to the skilled person that any such deviation can be weighted so that particular parts of the tone mapping curve have a weight higher than others.
It would also be possible to store tone mapping functions in tone mapping function table 7 in a way so that small changes of the look-up table-entry number correspond to small changes in the tone mapping function allows for a particularly simple way of determining whether or not the change from one tone mapping function of a given base frame to previous base frames is large or not. If the change in tone mapping function 6 is gradual, then the actual tone mapping function TMO* corresponds to the initial tone mapping function determined for base frame 4. If the changes of the tone mapping function determined for a given base frame 4 are too large, then the initial tone mapping function is corrected to a tone mapping function which corresponds closer to previous tone mapping functions. This can be determined by simple comparators. Accordingly, the change of tone mapping functions is small within a scene and thus the sequence of frames can be shown based on the compressed video data stream 12 which is of high quality.

Claims

Method of compressing a stream of video frame data wherein
tone mapping functions are determined for video frames of said stream,
said tone mapping functions being different from one another
for frames relating to different scenes
characterized in that
the tone mapping functions will be altered for frames of the stream relating to the same scene.
Method of compressing a stream of video frame data according to claim 1 wherein the stream or uncompressed video frame data is a stream of high dynamic range video frames.
Method of compressing a stream of video frame data according to any of the previous claims wherein the stream of uncompressed video frame data is a stream of high definition video data.
Method of compressing a stream of video frame data according to any of the previous claims wherein for at least some frames a base frame is determined.
Method of compressing a stream of video frame data according to the previous claim wherein a base frame is determined for a majority of frames
Method of compressing a stream of video frame data according to the previous claim wherein a base frame is determined for every frame.
Method of compressing a stream of video frame data according to any of the three previous claims wherein the tone mapping function is determined in response to a base frame.
Method of compressing a stream of video frame data according to any of the previous claims wherein a low dynamic range base frame determined in response to a base frame and information relating to a tone mapping function for the base frame is included in the stream of compressed video data.
Method of compressing a stream of video frame data according to any of the previous claims wherein the tone mapping function for a given frame is selected from one of a number of predefined tone mapping functions.
Method of compressing a stream of video frame data according to any of the previous claims wherein the predefined tone mapping functions comprise at least one fixed tone mapping function.
Method of compressing a stream of video frame data according to any of the previous claims wherein the predefined tone mapping functions comprise at least one user defined tone mapping function.
Method of compressing a stream of video frame data according to any of the previous claims wherein the low dynamic base frames are subjected to further base frame compression.
Method of compressing a stream of video frame data according to any of the previous claims wherein a frame detail is extracted from the frames of the uncompressed video data stream.
Method of compressing a stream of video frame data according to any of the previous claims wherein the frame detail is compressed separately from the base frame.
Method of compressing a stream of video frame data according to any of the previous claims wherein a tone mapping function is determined in response to information relating to a present and at least one previous frame.
Method of compressing a stream of video frame data according to the previous claims wherein the tone mapping function is altered gradually as long as no change of scene occurs.
Method of compressing a stream of video frame data according to any of the previous claims wherein a tone mapping function information is included in the video data stream separately from luminance information relating to the same frame.
Video camera for storing and/or transmitting a video data stream obtained according to a method of any of the previous method claims.
Video decoder for decoding a stream of video data obtained according to a method as described in any of the previous method claims.
A stream of video data compressed according to any of the previous method claims.
A data carrier carrying video data for providing a stream of video data according to claim 21.